基于分解–集成方法的铁路客运量预测

Railway Passenger Transportation Volume Prediction Models Based on Decomposition-Aggregation Methods

铁路客运量预测是铁路运输组织管理工作的重要基础和主要依据之一。本文建立多种分解–集成方法对全国铁路月度客运量进行预测分析。分别利用集合经验模态分解(EEMD)、奇异谱分解(SSA)和小波分解(WT)将原始序列进行分解,再分别使用季节差分移动自回归模型(SARIMA)模型和反向传播神经网络(BP)模型及其组合模型对分解后的子序列进行拟合、预测和集成。对比研究发现采用分解–集成方法有助于提高相关模型的预测准确性,且EEMD-SARIMA-BP组合模型在所有模型中预测效果最佳。

The forecast of railroad passenger volume is one of the important foundations and main bases of railroad transportation organization and management. In this paper, we establish multiple decom-position-aggregation methods to forecast and analyze the monthly passenger volume of national railroads. We use the ensemble empirical modal decomposition (EEMD), the singular spectrum de-composition (SSA) and the wavelet analysis (WT) respectively to decompose the original data series into several sub-series, then we process the forecast by fitting, forecasting and aggregating by the seasonal difference moving autoregressive model (SARIMA) model and back propagation neural network (BP) and their combined model of the sub-series, respectively. We find that the use of de-composition-aggregation methods helps to improve the prediction accuracy, and the combined EEMD-SARIMA-BP model has the best prediction effect among all models.